AI代理SDK

2. Install the SDK

npm install @pentatonic-ai/ai-agent-sdk

or: pip install pentatonic-ai-agent-sdk

```

Memory operations route through TES → engine. No client-side change between local and hosted.

to install a git post-commit hook so memory stays current

npx @pentatonic-ai/ai-agent-sdk onboard

Default .env points at Ollama on the host. Edit if your Ollama is

equivalent: npx @pentatonic-ai/ai-agent-sdk config hosted

Either way, verify with `/tes-memory:tes-status` in Claude Code, or from the shell:

The plugin's MCP server, hooks, and tools all read the same config — switching modes is a single CLI call away.

What it tracks (auto, every turn): - Memory search at prompt time — relevant memories injected as context - Memory store at turn end — every conversation turn persisted - Token usage — input, output, cache read, cache creation tokens per turn

`TESClient(config)` — Observability

`engineAdapter(config)` — Memory

Thin HTTP client for the memory engine. config = { engineUrl, arena, apiKey? }. Returns { ingestChunk(content, metadata), deleteByCorpusFile(repoAbs, relPath), init() }. See Use as a library.

For raw /store / /search calls, just fetch() against ${engineUrl} directly — the wire format is documented in packages/memory-engine-v2/MIGRATION.md.

---

What's in this SDK

Two products that share one TES account, one install line, and one dashboard:

Both products are sold separately, but you can use either, both, or neither. Plugins for Claude Code and OpenClaw install everything at once if you'd rather skip the SDK glue.

One-time: open browser, sign in or sign up, get API keys

npx @pentatonic-ai/ai-agent-sdk login

`login` opens your browser at the hosted sign-in page. New users click "Sign up" to create a tenant (clientId + region + email + password). After verification the CLI writes credentials to `~/.config/tes/credentials.json` (mode 0600). The Claude Code plugin, OpenClaw plugin, hooks, and corpus CLI all auto-discover this file — no manual paste step.

To check connection state later: npx @pentatonic-ai/ai-agent-sdk whoami. To point at a local TES dev instance: npx @pentatonic-ai/ai-agent-sdk login --endpoint http://localhost:8788.

(init still works as a one-major-release deprecation alias for login.)

---

What gets stored: references, not content

By default, ingest stores pointers to source content (path + line range + a short signature/summary), not full chunk content. Per-language strategies:

• Markdown — one reference per H1/H2 section
• JS / TS — one per top-level function / class / const / export
• Python — one per top-level def / class
• JSON / YAML — collapsed top-level keys
• Other — single file-level reference

Why pointers? Code mutates between ingests. Embedded chunks of old source rot silently — the LLM keeps confidently citing functions you've since rewritten, with retrieval evidence to back it up. Pointers rot loudly: when a file moves or changes, Read fails or returns different content, and the agent observes and adjusts. Stale-but-confident is the worst-class memory bug; loud-and-self-correcting is qualitatively better for source code.

It also means proprietary source never leaves your machine — only the index (path + summary) is sent to the hosted TES, and the agent reads actual file contents at query time on its own.

If you need a self-contained index (e.g. for air-gapped retrieval where the source isn't available at query time), opt into legacy chunk-content storage by passing mode: "content" to ingestCorpus when using the SDK as a library.

API Reference

Plugins

If you use Claude Code or OpenClaw, the plugin gives you both products at once — every conversation turn is captured (observability) AND searched/stored as memory. No SDK glue to write.

Plugins

Drop a .mjs file into ~/.config/pentatonic-ai/doctor-plugins/ to add your own checks. Useful for app-specific things — internal APIs, ingest freshness, custom infrastructure — without forking the SDK.

// ~/.config/pentatonic-ai/doctor-plugins/my-app.mjs
export default {
  name: "my-app",
  checks: [
    {
      name: "internal API",
      severity: "warning",
      run: async () => {
        const res = await fetch("https://internal/health");
        return res.ok
          ? { ok: true, msg: "200 OK" }
          : { ok: false, msg: `HTTP ${res.status}` };
      },
    },
  ],
};

See packages/doctor/README.md for the full plugin contract and programmatic API.

---